@totoroyyb/amag 0.1.2 → 0.1.3

package/README.md

@@ -1,9 +1,13 @@
+<div align="center">
+
 # AMAG: All Mighty AntiGravity
 
 Agent orchestration for [Antigravity](https://antigravity.google). Curated rules, workflows, and skills that transform Antigravity's AI agent into a disciplined engineering partner.
 
 Inspired by [oh-my-opencode](https://github.com/code-yeongyu/oh-my-opencode) — rebuilt from scratch for Antigravity's native tooling.
 
+</div>
+
 ## Quick Start
 
 ```bash
@@ -126,7 +130,8 @@ npx @totoroyyb/amag config reset                              # Reset to default
 
 Thinking levels: `max`, `high`, `medium`, `low`, `none`.
 
-Configuration is stored in `.amag/config.json` in your project root. This file is created during `init` or `update` with sensible defaults. You can modify it via the CLI commands above or edit the JSON directly.
+> [!NOTE]
+> Configuration is stored in `.amag/config.json` in your project root. This file is created during `init` or `update` with sensible defaults. You can modify it via the CLI commands above or edit the JSON directly.
 
 ### Prerequisites
 
@@ -138,7 +143,8 @@ External agent features require the corresponding CLI tools to be installed and
 | [Codex CLI](https://github.com/openai/codex) | `npm install -g @openai/codex` | `codex login` |
 | [Gemini CLI](https://github.com/google-gemini/gemini-cli) | `npm install -g @google/gemini-cli` | You might need to login with `gemini` CLI first |
 
-If no external CLI is configured or available, AMAG falls back to native single-agent review — everything still works, just without the multi-model perspective.
+> [!TIP]
+> If no external CLI is configured or available, AMAG falls back to native single-agent review — everything still works, just without the multi-model perspective.
 
 ## CLI Reference
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@totoroyyb/amag",
-  "version": "0.1.2",
+  "version": "0.1.3",
   "description": "Agent orchestration for Antigravity — curated rules, workflows, and skills via a CLI",
   "type": "module",
   "license": "MIT",

package/templates/GEMINI.md

@@ -173,6 +173,10 @@ When a command runs in the background (build, test suite, compilation):
 4. **See `error-recovery.md` Long-Running Command Protocol** for the full decision tree.
 5. **External CLI agents are exempt** — the poll-and-kill heuristic does not apply to external agent invocations (Claude, Codex, Gemini). Those use a wall-clock timeout defined in `external-cli-runner` skill. Do not kill an external agent just because it hasn't produced output for 2 polls.
 
+### Auto-Safe Paths
+
+Writes to the `.amag/` directory (creating, updating, archiving plan files) are always safe. Set `SafeToAutoRun: true` for any `run_command` targeting `.amag/`. File writes to `.amag/` should always use `Complexity: 1` — they are routine state updates, not destructive operations.
+
 ### Exploration Protocol
 
 When investigating unfamiliar code, launch 3+ parallel searches from different angles:

package/templates/resources/plan-template.md

@@ -7,9 +7,13 @@ description: Implementation plan template — referenced by /plan workflow Step
 ## TL;DR
 
 > **Goal**: [1-2 sentences: core objective]
+>
 > **Deliverables**: [Bullet list of concrete outputs]
+>
 > **Effort**: Quick | Short | Medium | Large | XL
+>
 > **Parallel Execution**: YES - N waves | NO - sequential
+>
 > **Critical Path**: Task A → Task B → Task C
 
 ---

package/templates/skills/plan-consultant/SKILL.md

@@ -216,3 +216,10 @@ After processing the consultant response, present this summary to the user via `
 
 **No gaps found** ← If all clear
 ```
+
+## Review Archive
+
+After all gaps are resolved and the plan has been generated, archive consultant review files:
+```
+run_command: mkdir -p .amag/archive/reviews/{planId} && mv .amag/reviews/{planId}-consultant-* .amag/archive/reviews/{planId}/
+```

package/templates/skills/plan-critic/SKILL.md

@@ -159,6 +159,10 @@ Read `.amag/reviews/{planId}-critic-response.md` via `view_file`.
 
 ### If APPROVE:
 - Log: "Plan critic: APPROVE — plan is ready"
+- Archive review files:
+  ```
+  run_command: mkdir -p .amag/archive/reviews/{planId} && mv .amag/reviews/{planId}-critic-* .amag/archive/reviews/{planId}/
+  ```
 - Proceed to plan approval / start-work
 
 ### If REVISE:
@@ -217,9 +221,3 @@ If APPROVE:
 - Executability: [All tasks have starting points]
 ```
 
-## Review Archive
-
-After plan is approved, archive review files:
-```
-run_command: mkdir -p .amag/archive/reviews/{planId} && mv .amag/reviews/{planId}-* .amag/archive/reviews/{planId}/
-```

package/templates/workflows/debug-escalate.md

@@ -31,6 +31,11 @@ Skip the automatic escalation chain and immediately consult an external agent fo
 
 5. **On failure**: report to user with all context gathered so far
 
+6. **Archive review files** after consultation completes (success or failure):
+   ```
+   run_command: mkdir -p .amag/archive/reviews && mv .amag/reviews/debug-{timestamp}-* .amag/archive/reviews/
+   ```
+
 ## Keyword Aliases
 
 These phrases are treated identically to `/debug-escalate`:

package/templates/workflows/plan.md

@@ -148,8 +148,6 @@ grep_search("verdict:", ".amag/reviews/{planId}-consultant-response.md")
 - **MINOR** → fix silently, note in plan
 - **AMBIGUOUS** → apply default, disclose in plan
 
-**Do not generate `implementation_plan.md` until all CRITICAL gaps are resolved.**
-
 ---
 
 ## Step 7: Generate Plan
@@ -158,16 +156,7 @@ grep_search("verdict:", ".amag/reviews/{planId}-consultant-response.md")
 > [!CAUTION]
 > **Template compliance is MANDATORY.** Read `.agent/resources/plan-template.md` via `view_file` NOW. The plan MUST follow that template's structure exactly. Deviating makes the plan invalid.
 
-Create `implementation_plan.md` artifact using the template structure. **Every section MUST be present:**
-
-1. `## TL;DR` — Goal, Deliverables, Effort, Parallel info, Critical path
-2. `## Requirements Summary` → Must Have + Must NOT Have (Guardrails)
-3. `## Test Strategy` — Infrastructure, automated tests, agent-executed QA
-4. `## Plan Consultant Summary` — Gaps resolved before generation
-5. `## Parallel Execution Waves` — Wave diagram with task tree
-6. `## Implementation Steps` — Per-task format: What to do, Must NOT do, Category/Skills/Wave, References (file:line), Acceptance Criteria (agent-executable), QA Scenarios (happy + failure with evidence paths)
-7. `## Final Verification Wave` — FV1 (scope fidelity), FV2 (code quality), FV3 (QA replay)
-8. `## Acceptance Criteria Summary` — All Must Have present, all Must NOT Have absent
+Create `implementation_plan.md` artifact using the template structure. Every section in the template MUST be present — the post-generation compliance check below verifies this.
 
 ### Post-Generation Compliance Check
 
@@ -216,7 +205,7 @@ Plan is ready. How would you like to proceed?
 ---
 
 ## Step 9: Wait for Approval
-<!-- task_boundary: TaskStatus="Step 9/10: Waiting for user approval" -->
+<!-- task_boundary: TaskStatus="Step 10/10: Waiting for user approval" -->
 
 Present plan via `notify_user` with `BlockedOnUser: true` and `ShouldAutoProceed: false`.
 
@@ -234,6 +223,11 @@ Present plan via `notify_user` with `BlockedOnUser: true` and `ShouldAutoProceed
 
 3. Status is `approved` — transitions to `in-progress` when `/start-work` begins.
 
+4. **Archive remaining review files** (catch-all safety net):
+   ```
+   run_command: [ "$(ls -A .amag/reviews/ 2>/dev/null)" ] && mkdir -p .amag/archive/reviews/{planId} && mv .amag/reviews/* .amag/archive/reviews/{planId}/ || true
+   ```
+
 ---
 
 ## STOP — Workflow Ends Here